It was broadly known that the addition of a small amount of sulfur into the reactor drastically changes the behavior of single-walled carbon nanotube (SWCNT) growth. However, due to the lack of in situ characterization technologies and the limitations of computational approaches, our understanding of the role of sulfur in SWCNT growth remains very poor. To resolve the long-term mystery of the carbon society, we employed a highly accurate machine learning force field (MLFF)-based molecular dynamics (MD) approach to explore the role of sulfur in SWCNT growth from Fe catalyst particles. We successfully grew defect-free SWCNTs on Fe catalyst particles with different sulfur contents via MLFF-MD simulations, and through systematic studies, we found that sulfur atoms are prone to passivate the surface of Fe clusters first. The sulfur-passivated Fe cluster surface is less active for carbon adsorption and SWCNT nucleation. Thus, the addition of both sulfur and carbon onto an Fe cluster during SWCNT growth leads to the formation of a sulfur-rich region and a carbon-rich region on the Fe catalyst surface, which facilitates the nucleation and growth of smaller SWCNTs from the carbon-rich regions. Consequently, more sulfur addition results in a smaller carbon-rich region and the growth of smaller SWCNTs, but an excessive amount of sulfur may poison the catalysts. This insightful understanding agrees very well with most experimental observations, and thus, the long-term mystery of the carbon society was successfully resolved by the artificial intelligence (AI)-assisted computational approach. These deep insights offer a strategy for synthesizing SWCNTs with controlled diameters through proper catalyst passivation, and they also show that the mechanism of SWCNT growth can be revealed via advanced theoretical studies powered by AI.